Anomalous event detection from videos using 3D convolutional network
P. S. Shanija, K. Rahamathulla · AIP conference proceedings · 2024
The automatic detection of unexpected events in surveillance data continues to worry researchers.Since maintaining public safety is the primary goal of installing video surveillance systems, monitoring the footage and responding quickly to it presents a big problem for people.Because anomalous events have a far lower probability of occurring than regular ones, there is also a huge loss of work and time.As a result, the adoption of an autonomous algorithm for abnormal event recognition has become crucial in video surveillance.Nevertheless, despite all of the research that has been done in this field, the automatic detection of anomalous events is still challenging or impossible because of a number of factors, such as environmental diversity, movement complexity, similarity between actions, crowded scenarios, etc.Despite how difficult it is to fix these problems; the biggest difficulty is locating enough identified anomalous event video recordings.More importantly, it can be difficult to find a substantial number of original videos that satisfy the aforementioned criteria.It can be challenging and time-consuming to distinguish between normal and abnormal behavior.This research presents a straightforward and practical framework for learning spatiotemporal attributes with deep 3-dimensional convolutional neural networks.